Bayesian document segmentation based on complex wavelet domain hidden Markov tree models

Abstract

A texture-based Bayesian document segmentation method is investigated in this paper. This Bayesian method is used to fuse texture likelihood and prior contextual knowledge to achieve document segmentation. The texture likelihood is based on a complex wavelet domain hidden Markov tree (HMT) model and the prior contextual is based on a hybrid tree model. A redundant wavelet domain Gaussian mixture model is employed to capture pixel-level features in the HMT model. Several document images are segmented to verify the proposed method. Comparisons with other corresponding models are made.